Assessing model misspecification in hypothesis testing with genetic variants for joint models of longitudinal and time-to-event data
Abstract
Many clinical studies collect longitudinal biomarkers that are strongly associated with time-to-event outcomes. Motivated by the challenge of testing genetic associations in this context, we explore joint models that link genetic variants with both longitudinal measurements and time-to-event data. In practice, the longitudinal traits are often nonlinear. To enhance robustness against potential misspecification caused by nonlinear trajectories in the longitudinal traits, we incorporate spline functions to capture subject-specific nonlinear evolutions. Empirical studies show that misspecification of the longitudinal trait model does not impact tests of SNP associations but does affect risk factor-survival associations. Finally, we apply the proposed methods to analyze data from the Diabetes Control and Complications Trial.
